Daily AI Briefing
The 6 top stories of the 120 AIBriefs curated that day.
OpenAI says an internal model significantly more capable than GPT-6 Astra produced the proof in 88 hours using ~10,000 coordinating AI agents, with a writeup and formal Lean proof. NYU's Tristan Buckmaster disputes the process, saying he and Levent Alpöge worked the problem for nearly a year using Claude and Codex.
Amodei's 3,500-word essay proposes slowing AI capability gains, with Anthropic unilaterally committing to give third-party evaluators permanent employee-level access. Altman agreed OpenAI will do the same; Musk, Hassabis and Nadella endorsed the direction.
OpenAI published first benchmarks for Jalapeño, its Broadcom-built inference chip: 1.5–1.9× more AI work per watt and 1.7–3.6× lower end-to-end latency than Nvidia GB200/GB300 systems. Tested on SemiAnalysis's InferenceX benchmark across GPT-OSS 120B, DeepSeek R1 670B and Kimi K2.5 1T.
OpenAI's GPT-6 Astra, billed as its most intelligent and aligned model, is behind a wave of community demos: a Monaco Grand Prix film rendered at 2,610 frames in 4K 30fps, a Blender-controlled VFX short, and a Minecraft run that collected 6 blaze rods and 3 ender pearls.
The open-weight multimodal MoE activates only 6B parameters per token and carries a 262,144-token native context, extensible to 1M with YaRN. Production Qwen3.8-Flash will hit the QwenCloud API at $0.16/1M input and $0.47/1M output tokens.
Trump called AI safety concerns a "hoax" and said his presidency is the only "guardrail" AI needs, after Anthropic CEO Dario Amodei's 3,822-word call to pace frontier development. A FINRA-style oversight body drafted by Treasury and OSTP was shelved after David Sacks and Mark Zuckerberg called Trump to object.
Also that day
102 more stories that day — open the live feed.
Get tomorrow's AI brief in your inbox